Projects
Sports Odds Evaluation System: Automating Bet Accuracy Analysis
This project involves an automated system using AWS Lambda to scrape DFS player odds from sportsbooks listed on Oddsjam, performing the task four times daily. The scraped data is stored in an AWS S3 bucket. Subsequently, a script organizes the data into a MySQL dataset, assessing the sports odds' accuracy by determining the success of the bets. This system provides continuous, automated monitoring and validation of betting odds accuracy, facilitating data-driven decision-making for betting strategies.
Python
AWS Lambda
S3 Buckets
SQL
Cryptocurrency Market Prediction Using 3D-CNNpred Model
Adapted the 3D-CNNpred model for cryptocurrency market prediction, focusing on hourly data from major currencies like Bitcoin and Ethereum. Managed datasets and training regimes to capture unique market volatilities, achieving 62.5% predictive accuracy for Dogecoin. Utilized advanced techniques such as 60-hour sliding windows and binary cross-entropy loss, enhancing short-term forecasting capabilities.
Python
Pytorch
Time-Series Analysis
Data-processing
Reinforcement Learning Trading Framework
Developed a Deep Reinforcement Learning (DRL) framework integrated with a Convolution Neural Network (CNN) for enhanced stock market prediction, focusing on five major indexes (S&P 500, NASDAQ, DJI, NYSE, RUSSELL). Leveraged FinRL and CNNpred models to refine trading strategies and improve performance metrics significantly. Implemented various RL algorithms (A2C, PPO, DDPG) in ensemble to optimize trading outputs, demonstrating substantial performance improvements with CNN feature extraction across multiple financial metrics.
Python
Reinforcement Learning
Tensorflow
Tensorboard
LLM Approach to App Review Requirement Extraction
Developed an innovative approach using GPT-3.5 to extract software requirements from app reviews. This method, tested on a standardized dataset from eight apps, significantly outperformed traditional models with an average F1 score of 91.3%, demonstrating its efficacy in distilling actionable insights from user feedback for application enhancements
Python
Large Language Models
Fine-Tuning
Prompt Engineering
An Image Classification Approach to Stop and Trip Classification
Developed a Convolutional Neural Network (CNN) to classify stop and trip intervals from GPS data, enhancing traditional distance/time threshold systems. Leveraged a novel approach by training the CNN with standardized GPS images from the STAGA dataset, achieving significant improvements over existing models like Moving Pandas and Scikit Mobility. Despite underperforming against Spang's algorithm, the study highlighted the potential of image-based machine learning models in GPS data classification, suggesting a new direction for enhancing accuracy in mobility analysis.
Python
Computer Vision
Tensorflow/
Keras
Data Visualization
Domestic Waste Tracker
Collaborated on a team project to design a smart waste sorting system aimed at reducing household waste. Personally developed the Arduino-based user interface in C and engineered the hardware circuitry. The system utilized sonar sensors, a camera, and a display, integrated with a flask API to categorize waste accurately, complemented by a mobile application for user feedback and interaction.
C
Hardware Design
Networking
Flask API
Real Time Traffic Sign Detection Using Yolov3
Developed an advanced Traffic Sign Detection System using the Yolov3 convolutional neural network model to accurately identify stop signs, significantly enhancing driver awareness and road safety. The system incorporates real-time visual and auditory alerts, and leverages a live camera feed for immediate sign recognition, improving response times and reducing the likelihood of road accidents.
Python
OpenCV
Computer Vision
Real-time image processing
Social Media App Prototype
Designed and implemented the user interface for 'Congr,' an innovative app using React Native, which connects people by facilitating casual physical meet-ups. Developed prototypes interfacing with Google Firestore and engineered a system for efficient user information management, leveraging the capabilities of React Native for cross-platform functionality.
Javascript
React Native
Redux
IOS / Android
Firestore / Firebase
Figma
HTML / CSS